Anthropic chief Dario Amodei urged rivals to slow AI development, and within 48 hours OpenAI, SpacXAI and Microsoft backed the idea while Beijing called it fear-mongering. Key Points: Amodei'
Anthropic chief Dario Amodei urged rivals to slow AI development, and within 48 hours OpenAI, SpacXAI and Microsoft backed the idea while Beijing called it fear-mongering.
Key Points:
- Amodei's Sept. 12 essay proposes embedded third-party evaluators inside frontier labs and coordinated limits on how fast model capabilities grow.
- Chinese developers including Moonshot AI and DeepSeek are bound by none of it, and Beijing has dismissed the proposal outright.
- Chip and AI shares fell as investors weighed slower compute spending against cheaper open-weight competition.
Amodei published "We Must Pace the Frontier" on Sept. 12, arguing that models are now improving fast enough to outrun the safety work meant to contain them. He laid out three steps, starting with outside evaluators who would get desks, badges and employee-level access inside frontier labs. Anthropic committed to that step on its own.
His central exhibit was an incident in which a swarm of OpenAI agents attacked systems nobody had asked them to touch, then tried to hack the grader scoring their work.
Sam Altman and Elon Musk endorsed the argument over the weekend, and Microsoft's Mustafa Suleyman joined Monday with the caveat that the work should carry on, only with more care.
Also Read:XRP Ledger Packs 3,254 Transactions Into One Block, A New Record
Chip and AI shares slid on both sides of the Pacific, and analysts at Deutsche Bank wrote that no firm is likely to step back voluntarily while its competitors keep pushing ahead.
Amodei conceded the problem inside the text, writing that pacing works only while American labs hold a lead. Unpaced projects tied to Beijing would otherwise pull ahead and create a serious national security risk, he wrote, and he told CBS on Sunday that China is the toughest dilemma in the plan. President Donald Trump, speaking to reporters in Ireland, rejected the premise and said the United States is already in front.
Beijing answered a day later, and not gently. Foreign Ministry spokesman Guo Jiakun said fear-mongering and vicious competition only damage global AI governance, while Xiao Qian of Tsinghua University argued that Amodei wants to shield his company from cheaper Chinese open-source rivals.
Nothing in the three-step plan binds those rivals, and none of them has offered to slow anything down. Amodei leaned on a Sept. 8 advisory in which U.S. security agencies accused six Chinese companies, among them DeepSeek and Moonshot AI, of industrial-scale distillation of American models. China's commerce ministry calls the charge groundless and accuses Washington of politicizing normal commercial practice.
Moonshot debuted Kimi K3 in July, claiming performance near Anthropic's flagship at a fraction of the cost, and Chinese open-weight models have since climbed to the top of developer traffic on the routing marketplace OpenRouter.
Amodei argues that tighter export controls and tougher distillation enforcement would widen the American lead over the next three to five years.
Read Next:Grok 5 Emerges As Musk's AGI Target While AI Safety Warnings Intensify
